Explaining the “Identiﬁable Victim Effect” · explaining the “identifiable victim effect”

JOBNAME: Kluwer Journals − RI PAGE: 1 SESS: 34 OUTPUT: Wed May 28 08:52:24 1997/data11/kluwer/journals/risk/v14n3art2

Journal of Risk and Uncertainty, 14:235–257 (1997)© 1997 Kluwer Academic Publishers

Explaining the “Identifiable Victim Effect”

KAREN E. JENNIDepartment of Engineering and Public Policy, Carnegie Mellon University

GEORGE LOEWENSTEINDepartment of Social and Decision Sciences, Carnegie Mellon University

Abstract

It is widely believed that people are willing to expend greater resources to save the lives of identified victimsthan to save equal numbers of unidentified or statistical victims. There are many possible causes of this disparitywhich have not been enumerated previously or tested empirically. We discuss four possible causes of the“identifiable victim effect” and present the results of two studies which indicate that the most important causeof the disparity in treatment of identifiable and statistical lives is that, for identifiable victims, a high proportionof those at risk can be saved.

Key words: value of life, identifiable victims

JEL Classification: J-17

“There is a distinction between an individual life and a statistical life. Let a 6-year-oldgirl with brown hair need thousands of dollars for an operation that will prolong herlife until Christmas, and the post office will be swamped with nickels and dimes to saveher. But let it be reported that without a sales tax the hospital facilities of Massachu-setts will deteriorate and cause a barely perceptible increase in preventable deaths—notmany will drop a tear or reach for their checkbooks.” (Schelling, 1968)

“The death of a single Russion soldier is a tragedy. A million deaths is a statistic.”Joseph Stalin (quoted in Nisbett and Ross, 1980:43)

In late 1987, eighteen-month old Jessica McClure spent 58 hours trapped in a well, andAmericans responded with sympathy, a tremendous rescue effort, and money. The Mc-Clures received over $700,000 in donations for “baby Jessica” in the months after herrescue, and eventually a popular television movie, “Everybody’s Baby: The Rescue ofJessica McClure,” was made about the incident (People Weekly, November 2, 1987; April16, 1990; Variety, May 31, 1989). At the time, there was no question but that everythingpossible should and would be done to rescue the child; cost was no object. If similarresources were spent on preventative health care for children, hundreds of lives could besaved. Yet it is difficult to raise money for efforts directed at saving such “statistical”victims.

Kluwer Journal@ats-ss12/data11/kluwer/journals/risk/v14n3art2 COMPOSED: 03/25/97 10:14 am. PG.POS. 1 SESSION: 13

5% 50% 90% 100%


The story of “baby Jessica” is simply one example of the “identifiable victim effect:”society is willing to spend far more money to save the lives of identifiable victims than tosave statistical victims. This has been remarked upon in treatises on public policy (Gore,1992), in scholarly works (Schelling, 1968; Calabresi and Bobbitt, 1978; Viscusi, 1992;Whipple, 1992), the medical literature (Redelmeier and Tversky, 1990) and the popularpress (Toufexis, 1993).

The identifiable victim effect plays a role in many important policy issues. Recently, ithas received special prominence in the national debate over funding priorities for healthcare, where expensive measures are often taken for identified individuals, but funding forpreventative care seems to be lacking. For example, a recent effort to separate conjoinedtwins, whose probability of surviving the operation was estimated to be less than 1%, wasused in some media to highlight the discrepancies between extravagant health care fund-ing for “last-ditch efforts … to save the few” and modest funding for basic and preven-tative care that would benefit the many (Toufexis, 1993). In debates over the NorthAmerican Free Trade Agreement, opponents could identify specific individuals whowould lose their jobs if the agreement was approved, whereas proponents could refer onlyto the additional “statistical” jobs that would presumably result (Goodman, 1993). Iden-tifiable victims need not be human: in 1988 a multi-national effort spent millions to rescuethree grey whales trapped under the Arctic ice cap, while at the same time the Japanesewhaling industry was spending millions to locate and harvest whales (Linden, 1988).

Identifiable victims seem to produce a greater empathic response, accompanied bygreater willingness to make personal sacrifices to provide aid. One might think, therefore,that the large literature on empathy, altruism, and helping behavior would provide cluesabout why identifiable victims are treated differently from statistical victims. However, theliterature on helping behavior focuses almost exclusively on the factors that cause peopleto aid identified victims (see, e.g., Latané and Darley, 1970, or Piliavin et al., 1981), andmuch of this literature looks at factors, such as the number of potential aiders and thecosts of providing aid, that are not obviously relevant to the problem of why identifiableand statistical victims are treated differently. Likewise, the literature on empathy andaltruism has been concerned primarily with the question of whether “true”—that is self-less—empathy actually exists (see, e.g., Cialdini et al., 1987, and Batson et al., 1991),which again seems to have little relevance to the question of why identifiable and statis-tical victims produce such a different response. We have not seen any explicit treatment ofthe identifiable victim effect in either literature.

In those literatures where it has been discussed, the distinction between identifiable andstatistical victims is typically treated as a simple dichotomy, and the frequency with whichit is mentioned reinforces this view. However, the simplicity of the distinction is deceptive:in practice, there are several differences between identifiable and statistical victims, anyone of which could account for their differential treatment.

Our goal in this paper is to gain a better understanding of the psychological underpin-nings of the identifiable victim effect. We do not attempt to explain the effect at a deeperlevel—e.g., to explain at an evolutionary level how or why humans have come to respondmore strongly to identifiable than to statistical victims. Based on discussions with col-

236 JENNI/LOEWENSTEIN


5% 50% 90% 100%


leagues and a combing of the literature, we came up with four factors that differ betweenstatistical and identifiable victims that could potentially account for their differentialtreatment. Although we do find that one of these factors appears to be an important causeof the effect, it is possible that other factors we have not identified also play a role.

In what follows, we first discuss these four differences between identifiable and statis-tical victims which may be responsible for the effect. We then discuss the normative statusof the effect in relation to each of the four possible causes. Finally, we present findingsfrom two studies designed to test the four possible causes and to determine whether, ifsupported, they are consciously endorsed.

1. Potential causes of the identifiable victim effect

1.1. Vividness

When an identifiable person is at risk of death, the media tell us a lot about them, and wemay come to feel that we know them1. Research on “vividness” has shown that specific,concrete examples have far greater influence on what people think and how they behavethan more comprehensive but pallid statistical information (Nisbett and Ross, 1980).Situations with identifiable victims are often characterized by all the major factors thatconvey vividness: the stories are very emotional (victims featured in the media are oftenparticularly sympathetic, helpless, or blameless), we see visual images of the victim innewspapers and on television, and we see the events unfold in real-time—without theemotional distance provided by a historical perspective. For example, we see the pictureof the small girl who is trapped in the well, interviews with her tearful parents ontelevision, and live coverage of the desperate attempt to rescue her. These vivid detailsmay result in a perceived familiarity with the victim, making it seem more important toundertake extraordinary measures to save that person. As Schelling (1968) expresses it,“the more we know, the more we care.”

Indeed, many public relations and marketing tactics seem to be premised on the viewthat the vividness of an identifiable victim will enhance the public’s desire to “do some-thing” about the problem. For example, the “poster child” for MS fund raising, and thepictures and life stories that accompany requests for money to prevent malnutrition pointto a widespread belief that concrete details increase the public’s concern. Likewise,arguments for and against the proposed “three strikes and you’re out” federal sentencingpolicy rely on vividness to create sympathy for their position: arguments for implementingsuch a system discuss specific victims of violent crimes who would not have been vic-timized had the “three strikes” policy been in place (Skelton, 1993), whereas argumentsagainst the policy focus on relatively harmless three-time offenders who would facelifetime incarceration (Egan, 1994).

EXPLAINING THE “IDENTIFIABLE VICTIM EFFECT” 237


5% 50% 90% 100%


1.2. Certainty and uncertainty

A second distinction between identifiable and statistical victims is that identifiable deathsare usually certain to occur if action is not taken, whereas statistical deaths, by definition,are probabilistic. Since they are certain to occur if action is not taken, the subjectiveimportance of identifiable deaths may be enhanced by the “certainty effect”—the ten-dency to place disproportionate weight on outcomes that are certain relative to those thatare uncertain but likely (Kahneman and Tversky, 1979). In addition, there is compellingevidence that people are typically risk-seeking for losses (Kahneman and Tversky, 1979;Tversky and Kahneman, 1981, 1986; Cohen, Jaffrey, and Said, 1987). Risk-seeking forlosses implies that a certain loss is seen as worse than an uncertain loss with the sameexpected value. For example, most people prefer a 50;50 chance of losing $100 or losingnothing to a certain loss of $50. Risk-seeking for losses has been demonstrated fornon-monetary losses as well: in one well-known study (Tversky and Kahneman, 1981),subjects were given two identical scenarios in which lives were at risk and were asked tochoose between two treatment options. In one case, the scenarios were worded in terms oflives saved (gains), and in the other they were worded in terms of lives lost (losses).Consistent with the prediction that people are risk-seeking for losses, most subjects in thelives lost condition chose the riskier treatment option. Risk-seeking for losses implies thatthe number of certain (identifiable) fatalities that is deemed “equivalent” to uncertain(statistical) fatalities is less than the expected number of statistical deaths. Both thecertainty effect and risk seeking for losses, therefore, may contribute to the tendency totreat identifiable (and thus certain) victims as more worthy of attention than statisticalvictims.

1.3. Proportion of the reference group that can be saved

Public perceptions of risk are responsive to the distribution of risk among the populationas well as to the absolute level of risk (Slovic, Fischhoff, and Lichtenstein, 1980). Ingeneral people are more concerned about risks that are concentrated within a geographicregion or population than about those that are dispersed (National Research Council,1989). This concern with the concentration of risk may help to explain the identifiablevictim effect: identifiable victims represent highly concentrated distributions of riskwithin a specific reference group. In effect, identifiable victims become their own refer-ence group, creating a situation where n out of n people will die if action is not taken. Forexample, if 120 people are likely to die in a plane crash this year, these are only 120people out of the millions who fly. Once a plane carrying 120 passengers crashes with allaboard lost, however, these are 120 fatalities out of the 120 on board the plane.

There is considerable evidence that, holding the number of victims constant, people’sconcern increases as the applicable reference group shifts. For example, people are lesstolerant of the risks of vaccination when there is a smaller “risk group” for vaccine sideeffects, even when members of that risk group could not be identified a priori (Ritov andBaron, 1990). Similarly, economic studies of the “value of life” have found that the value



5% 50% 90% 100%


of avoiding death or injury increases as the baseline probability of death or injury in-creases (Weinstein, Shepard, and Pliskin, 1980; Viscusi and Evans, 1990; Horowitz andCarson, 1993). Willingness to pay for small reductions in risk can be extrapolated tocalculate a value of life. For example, the value of life can be calculated as 10 times thewillingness to pay to avoid a 10% chance of death, or as 100 times the willingness to payto avoid a 1% chance of death. The willingness to pay to avoid a 10% chance of death isgreater than ten times the willingness to pay to avoid a 1% chance of death, resulting ina higher overall value of life when the baseline risk is high. Since a high proportion of thereference group at risk implies a high probability of fatality, or high baseline risk, for eachmember of the risk group (e.g., if 25 out of 100 will die the baseline risk for each memberof the reference group is 0.25, but if 25 out of 50,000 will die, the baseline risk for eachmember of the reference group is 0.0005), these findings are consistent with an increasein concern about fatalities when the reference group is small relative to the number at risk.

According to the proportion of the reference group at risk explanation, there is not astrict dichotomy between identifiable and statistical lives. Instead, identifiable victims lieat one end of a continuum running from low probability risks spread over the entirepopulation (statistical deaths) to certain death for every member of the population (iden-tifiable deaths).

1.4. Ex post versus ex ante evaluation

Identifiable victims are actual people who are very likely to die or be injured, whereasstatistical victims are, as the term implies, simply statistics. In other words, with identi-fiable victims, both they and we know, at the time we have to decide what to do, that theyare likely to die as the result of a preventable or addressable cause. The decision aboutrescuing an identifiable victim, or the evaluation of the value of rescuing the victim, isusually made ex post, or after, the occurrence of some risk-producing event. In contrast,the evaluation of the value of addressing risks to statistical victims is usually made exante, or before the risk-producing event has occurred. (Weinstein, Shepard, and Pliskin,1980). The ex post/ex ante distinction appears to be the identifiable victim effect itself,after other factors that covary closely with the identifiable/statistical discrepancy—i.e.,vividness, certainty, and the proportion of the reference group at risk—have been elimi-nated.

Ex post evaluation makes it more difficult to apply cost-benefit principles in decidingwhat to do, and instead makes issues of responsibility and blame salient. Once a victimhas been identified ex post, people can no longer “withdraw to a naked statistical analysisof the cost-effectiveness of the effort,” whereas people have few reservations about doingso ex ante (Gillette and Hopkins, 1988). In addition, peoples’ perceptions and judgmentsof “risk” depend in part on the saliency of blame (Douglas, 1992). The possibility for therecognition of responsibility, and thus the attribution of blame, is clear in the ex post caseand almost absent in the ex ante case.



5% 50% 90% 100%


2. Normative status of the identifiable victim effect

Scholars are divided about the identifiable victim effect’s normative status. Emphasis onsaving identifiable victims has been deplored as irrational (MacLean, 1986; Whipple,1992), and praised as humanizing (Glover, 1977; Calabresi and Bobbitt, 1978; Gibbard,1986; Gillette and Hopkins, 1988). For example, MacLean (1986) argues that activitiesundertaken to rescue identifiable victims, when compared to the efforts spent to reducestatistical risks, “defy economic or even risk-minimizing sense,” whereas Gibbard (1986)asserts that it is immoral not to act when identifiable lives can be saved. Although thenormative status of the effect may not be important for individual decision making, it isextremely important when identifiable victims are used to justify or defend specific policydecisions. How should the sympathy we feel for identifiable victims affect public policy?

The normative defensibility of the effect depends, in part, on its cause. For example,since people generally obtain more information about, and see more vivid descriptions of,identifiable victims than statistical victims, identified victims may be seen as more famil-iar. Although people might reasonably respond in a more emotional fashion to familiar orvivid victims, it is less reasonable to endorse a policy that gives higher priority to morefamiliar victims, all else held equal, than to less familiar victims. This view would amountto allowing media coverage to determine aid allocation.

The normative status of the effect of certainty lies in a grey area. Although many peopledisplay an analogous pattern of behavior when deciding between certain and uncertainmonetary losses, it is not clear whether risk seeking for losses is a principle that peopledo, or should, consciously endorse or, if so, whether that the same principle should applyto decisions involving lives. Indeed, it is always possible to reframe a decision involvingtradeoffs between numbers of deaths as one involving saving lives.

There may be normative arguments for being concerned with the distribution as well asthe absolute magnitude of potential harm. MacLean (1986) believes that we should “dis-tribute risks of death equally or in proportion to the distribution of expected benefits.”Along similar lines, several moral philosophers argue that “fairness” or equity is a criticalfactor both in defining justice and in evaluating risks (Rawls, 1971, 1993; Shrader-Frechette, 1991; Raynor, 1992), a view which is especially cogent in light of the envi-ronmental justice movement, and the claim that the risks of hazardous waste disposal andpollution are borne disproportionately by minorities and by the poor (Bullard, 1993;Cushman, 1994).

However, in other situations taking the risk distribution into account seems less defen-sible. Given that reference group size is often a matter of framing—a reference group ofarbitrary size can be specified for virtually any hazard—a blanket endorsement of a policythat treats fatalities differently based on what proportion of the reference group theycompose is normatively dubious. For example, it probably makes no sense to treat adisease that kills 100% of the 10% of the population susceptible to it differently from onethat kills 10% of the 100% of the population susceptible to it. However, some referencegroups may be more normatively defensible than others. Thus, even after careful consid-eration, one might be more upset about a disease that kills an entire family or people in



5% 50% 90% 100%


a small geographic area than one that kills a similar number of victims from around thecountry.

The normative status of the distinction between ex post and ex ante evaluation of risksis also ambiguous. Although ex post evaluation of identifiable victims may bring into playpowerful emotions that do not apply to statistical victims, it is not clear what role thosereactions should play in making policy decisions.

These four possible causes of the identifiable victim effect are, it seems, differentiallydefensible. Thus shedding light on which, if any, of these causes are responsible for theeffect will also help to determine its normative status. In addition, it is not clear whetherthe identifiable victim effect is the result of a reasoned response, or if it is a gut-level,instinctive response. Whatever the cause of the identifiable victim effect, it would beinteresting to ascertain whether people embrace it as a principle of decision making. Thatis, do we value identifiable lives more than statistical lives unconsciously and possiblyunintentionally, or do we continue to value identifiable lives more after reflection? Theanswer to this question may also reflect on the normative status of the effect.

3. The studies

Our first goal was to determine which of these four potential causes, if any, contribute tothe identifiable victim effect. In both studies, subjects read risk and accident scenarios inwhich each cause was either present or absent, and then rated the importance of saving thevictim(s)2. In the first study we had a second goal: to determine if people continue todistinguish between identifiable and statistical victims when faced with an explicit choicebetween saving identifiable versus statistical lives. To address this question, we includedtwo judgment conditions: rating and direct comparisons.

3.1. Study 1

Method. To test the various explanations for the identifiable victim effect, and to deter-mine whether, if supported, they are consciously endorsed, we developed sets of scenariosin which the four causal factors were manipulated. We asked subjects to judge the im-portance of reducing risks in each, by having them either rate or compare the scenarios.

To investigate whether the identifiable victim effect is unconscious, or whether it per-sists in the face of obvious comparisons of identifiable and statistical victims, we testedsubjects in two conditions. In the rating condition, subjects saw the scenarios in randomorder, with scenarios related to the same explanation separated by other questions. Theyrated the importance of eliminating risks in each scenario (compared to other risks forwhich the government has responsibility) on a one-to-five scale, where 1 was “not de-serving of attention,” and 5 was “one of the most important.” In the direct comparisoncondition, subjects read the scenarios designed to test a single explanation together, andchose the situation in which it was more important to eliminate the risk. “Equally impor-tant” was given as an option. The direct comparison version makes it clear to subjects that



5% 50% 90% 100%


the same number of lives could be saved in each case, and makes the manipulation highlysalient. Given this salience, we can assume that any choice other than indifference is theresult of a conscious judgment on the part of subjects about the importance of thedistinction. In the rating condition, on the other hand, subjects were unaware of what wasspecial about the scenarios they viewed. For example, those who were exposed to a victimdescribed in detail to increase vividness were not aware that there was another conditionin which details were not provided. Thus, they would have a much harder time avoidinga gut-level response, even if they believed that normatively it should not affect theirjudgments.

If a particular effect operates unconsciously and unintentionally, differences caused bythe manipulation should be evident in the rating condition but not in the direct comparisoncondition. Alternatively, if the asymmetry is consciously endorsed, we would expect it tobe equally strong in both the rating and the comparison groups, or possibly even strongerin the latter.

To test whether vividness increases the apparent importance of undertaking a rescueattempt, we presented two scenarios involving a traffic accident in which a victim wasinjured and required immediate, possibly expensive, medical help. The two scenarios arepresented in the first panel of Table 1. The two situations are identical except that in onecase no information about the victim was provided, and in the other case we provided abrief description of the victim. If vividness is a critical factor in making decisions aboutlife-saving actions, subjects should rate it as more important to save the vividly describedvictim than the anonymous victim.

Testing the effect of uncertainty is complicated by the fact that (at least) two types ofuncertainty might be relevant—uncertainty about whether a specific individual will be-come a victim, and uncertainty about whether that individual, once a victim, will die. Inthis first study, the effect of certainty versus uncertainty was examined with two scenariosin which the expected number of fatalities from a contaminated food source was heldconstant (at ten), but in one case the deaths are certain to occur if action is not taken, andin one case they are probabilistic. In the probabilistic case, the scenario explicitly statedthat fewer or more than ten may die. These scenarios are shown in Table 2. If people arerisk seeking when it comes to deaths, we would expect most subjects to find it moreimportant to prevent the ten certain deaths than to prevent the ten probabilistic deaths.

To test whether the proportion of the risk group that can be saved affects subjects’preferences for risk-reducing projects, we developed two scenarios involving traffic fa-talities and different at-risk populations. Table 3 presents these two scenarios. In both,“exactly” 25 lives could be saved by a new safety program, but the number of peoplespecified as being at risk was 50,000 in one case and 25 in the other. According to theproportion-of-the-reference-group hypothesis, the problem should be seen as more severeas the percentage of the reference group at risk increases, so that saving 25 out of 25 atrisk will be considered more important than saving 25 out of 50,000.

Finally, to test whether people believe that ex post decisions to save lives are moreimportant than ex ante decisions to protect them, we developed two parallel scenarioswhere one individual was at risk from a pesticide being field-tested (see Table 4). Thescenarios differed only in whether action could be taken before or after the pesticide had



5% 50% 90% 100%


Table 1. Scenarios testing the effect of vividness.

Study 1

[Anonymous] There has been a traffic accident on a remote section of the highway, and a person has beenseriously injured. This person requires a helicopter rescue and immediate medical treatment to save his life.

[Vivid] There has been a traffic accident on a remote section of the highway, and a young secretary hasbeen seriously injured. The secretary was traveling by herself, on her way to spend the weekend with herparents. She requires a helicopter rescue and immediate medical treatment to save her life.

Study 2

The 1994 earthquake in Southern California caused approximately 34 deaths, thousands of injuries, overa billion dollars of structural damage to buildings, and seriously damaged six of the major freeways in theregion. Two days after the earthquake, rescue workers discovered a victim trapped in a collapsed parking garagelocated very near Highway 17. The news story reproduced below describes the rescue efforts that were under-taken at the time.

Example of news story with a “described” victimRescue Workers Find Quake SurvivorEfforts to remove survivor underway

Los Angeles, Calif. Jan. 15.Rescue workers were losing hope.It’s not possible, not in thatwreckage. There couldn’t be any-one alive in there.

A few still hoped. Perhaps,they thought, as they probed thewreckage of the municipal park-ing garage at the 23rd Ave. exit ofHighway 17 with high-tech survi-vor, and are working as quickly aspossible to sort through the rubbleto reach him.

Don Grisom, the attendant atthe parking garage, rememberswaving to Bob Wright as he wentto his car shortly before 3 pm.Then the earthquake struck andthe garage collapsed. Mr. Grisomwas able to run to safety, but sawMr. Wright just getting into hiscar as the building began to comedown. Mr. Wright, 42, is a localhigh school teacher and basketballcoach. His wife, Mary, has notleft the equipment, perhaps some-

one could have survived the col-lapse.

Finally, just after 6 am, aworker using a special fiber opticcamera to search parts of the ga-rage that are currently inaccessiblespotted a hand moving in the win-dow of a car. Someone was alivein that mountain of rubble—im-prisoned in a tomb of concrete,but alive nonetheless. Scene sinceshe heard her husband might betrapped inside. She is ecstatic thathe has been found, but is veryconcerned. “It’s frightening,” shesays, “But Bob is strong and hasan incredible zest for life. I knowhe will hold on for me and forJimmy [the couple’s 3 year oldson]. I am hoping and praying forhim, and I’m sure he is going tobe OK.”

While rescue workers arecautiously optimistic, the survivorappears to be pinned in his car,underneath tons of concrete and

Search of the partially col-lapsed parking structure beganyesterday afternoon, after a park-ing attendant, who miraculouslyescaped the collapsing garage dur-ing the earthquake, rememberedseeing a man walk into the garageto pick up his car at about 3 pm,minutes before the quake hit.

Workers and paramedics havemade voice contact with the steelrubble, and in a particularly un-stable section of the collapsedstructure. Before full-scale extrac-tion measures can begin, parts ofthe building must be stabilized.City engineers are on the site di-recting the stabilizing work, butare unsure how long it will bebefore they will be able to removethe survivor. However, rescuersare also working to create an ac-cess route so paramedics canreach the survivor quickly andattend to any critical injuries.



5% 50% 90% 100%


been applied. In the ex post case the individual had been exposed and had to be located;in the ex ante case the individual was about to be exposed, unless she could be locatedfirst. The ex post/ex ante distinction suggests that the former scenario will be judged moreimportant.

Table 1. (Continued)

What isn’t in this news story is a controversy over the effect of nearby freeway traffic on the parking

structure. Highway 17 was not damaged by the earthquake, and was a primary alternative route for all traffic

traveling from the south into downtown Los Angeles. Consequently, the highway was carrying much more

traffic than usual. The highway passes directly adjacent to the parking garage, and many engineers were

concerned that vibrations from the traffic on the highway could cause more damage to the parking garage,

further endangering the trapped victim. These engineers suggested shutting down the highway until the victim

could be extracted from the collapsed garage. However, the highway was estimated to be carrying about 230,000

cars per day, all of which would be diverted onto surface streets if it were shut down. This would add an average

of an hour an a half to commuting time into downtown. Keeping the highway open would not endanger the

rescuers, but could produce further structural shifting, with risks to the trapped victim. If you had been in the

position of having to make a decision about the freeway, how strongly would you have supported closing the

freeway?

Table 2. Scenarios testing the effect of certainty and uncertainty.

Study 1

[Uncertain] A major food distributor has just discovered that its newly introduced yogurt can cause deathto individuals who are allergic to its new ingredient. Approximately 1% of the general population is allergic tothis particular ingredient, but the existence of the allergy is not well known. The yogurt has been distributed toa large number of retailers, and sold to about 1000 different consumers. Each consumer therefore has a 1%chance of becoming ill and dying from the yogurt, and the best estimate is that 10 people will die, but more orfewer may die depending on the prevalence of the allergy. These people can be saved if all the consumers arelocated and treated.

[Certain] A major food distributor has just discovered that a small number (10 containers) of the yogurtsit distributed yesterday were contaminated, and that the contamination will result in the death of anyone who haseaten the yogurt, unless an antidote drug is administered. The yogurt was distributed to a large number ofretailers, and has since been sold to approximately 1000 consumers. 10 deaths will result unless the consumersare located and the antidote drug administered. If all the consumers are located and the antidote is administered,no one will be harmed.

Study 2

On average, about 100 children under the age of 4 die each year from suffocation associated with thin-layerplastic (bags and wrappings) in the United States. Over the past 10 years, annual deaths have ranged from [92to 110] [34 to 168]. Most of these deaths are associated with plastic bags used by dry cleaners and with plasticwrapping from toy packages. Some scientists have suggested that replacing plastic with paper wrappings forthese purposes would virtually eliminate the problem of children suffocating on plastic wrap. Legislation hasbeen introduced that will require all dry cleaners and toy manufacturers in the U.S. to begin packaging withpaper rather than plastic in 1996, with complete phase-out of plastic by the year 2000. Think about otherlegislation concerned with safety that is being, or could be, considered, from legislation you don’t care at allabout to legislation you care very much about. Relative to other legislation, what priority would you place onhaving this legislation passed and implemented?



5% 50% 90% 100%


These scenarios may seem, in isolation, to be artificial and unrealistic. This is in partbecause real-world identifiable victims are often characterized by several of these distinc-tions, so it is extremely difficult to manipulate them independently in a complete factorialdesign.

The existence of any one of the four factors implies the existance of at least one otherfactor. For example, if a victim is certain, that victim also composes a large proportion ofthe reference group (in this case, certain victims automatically become their own refer-ence group, and in fact, compose 100% of the reference group). However, when a largeportion of the reference group is at risk, those victims do not have to be certain victims.For example, everyone in a population could be susceptible to a specific disease, but theprobability that the disease will kill a susceptible individual might be quite low.

Figure 1 illustrates the relationships between the four factors, where arrows representlogical implication. For example, the arrow from certain to high proportion of referencegroup indicates that a certain victim also represents a high proportion of the referencegroup. However, there is no arrow from high proportion to certain, so victims whocomprise a high proportion of the reference group are not necessarily certain victims.

The complex relationships between these four possible causes make it impossible todevelop scenarios in which all four causes are varied independently. For example, if factorA is necessary for factor B, the combination of B, not A, is logically impossible. However,no one factor is both necessary and sufficient for any other factor, and thus no two of thesefactors are perfectly confounded. This makes it possible to examine the effect of eachcause independently, with the other three held constant, as we did in these two studies.

Table 3. Scenarios testing the effect of the proportion of the reference group that can be saved.

Study 1

[Large reference group] Approximately 50,000 people die every year in traffic accidents in the UnitedStates. A new program has been proposed that will save exactly 25 of these 50,000 lives every year.

[Small reference group] 25 people die every year in traffic accidents on a specific highway interchange. Anew program has been proposed that will eliminate the risks at this interchange. If the program is undertaken,it is expected that there will be no further fatalities at this interchange.

Study 2

At an intersection in downtown Pittsburgh, there are several automobile-pedestrian accidents every year,and in each of the past three years 2 pedestrians have been killed at that intersection. These pedestrian accidentsaccount for [2 of 4 people who died in accidents at that intersection] [2 of 112 people who died in auto-relatedaccidents in southwestern Pennsylvania] [2 of 1700 people who died in auto-related accidents in Pennsylvania]last year. To reduce the risks to pedestrians, city engineers have proposed installing automobile barriers betweenthe sidewalks and the streets surrounding this intersection, and building a pedestrian overpass. Althoughexpensive, they believe these measures will eliminate the possibility of future pedestrian accidents at theintersection. At the same time, there are many other useful traffic safety and improvement projects proposedboth for the city and for the state, and there is not enough funding to implement them all. Relative to otherautomobile and traffic safety projects you know about or can imagine, what priority do you think should begiven to funding this project?



5% 50% 90% 100%


Subjects. The questionnaire was administered to 70 undergraduates at Carnegie MellonUniversity, 30 visitors to a mall in south suburban Boston, and 27 visitors (mainly stu-dents) to the University of Pittsburgh and Carnegie Mellon University student centers.Forty-one respondents received the direct comparison questionnaire, and 86 received therating version. We tested approximately twice as many subjects in the rating conditionbecause the statistical power of the planned between-subjects comparisons is lower than

Table 4. Scenarios testing the effect of ex post and ex ante evaluation.

Study 1

[Ex post] The EPA has approved a field test of a new pesticide and it has been applied to a wheat field. Ithas just been discovered that someone from a nearby community was exposed to toxic levels of the pesticideduring the application and that if they are not treated quickly, they will die. Locating this individual will requireimmediate door-to-door search of the local community.

[Ex ante] The EPA has approved a field test of a new pesticide and is about to apply it to a wheat field. Ithas just been discovered that someone from a nearby community is camping near the field and will soon beexposed to toxic levels of the pesticide, such that they will die from the exposure. To prevent this exposure willrequire conducting a thorough search of the nearby area.

Study 2

Park rangers for the California Department of Parks have just closed a small area in the remote SierraNevada to camping and hiking, due to the discovery of strong toxic chemicals in a fresh-water spring locatednear an abandoned mine. Campers in that region are required to sign in and sign out with park officials, so allcampers are now being warned to steer clear of this abandoned mine. However, in checking the filed hikingplans of visitors currently in the general area, they discover that a hiker already on the trail has plans to campin the now-closed area near the abandoned mine. They suspect he intends to use water from the springs forcooking and drinking. According to the hiker’s plans, he [will enter the contaminated area tomorrow] [enteredthe contaminated area yesterday]. Rangers have the option of asking the local search and rescue team to go afterthe hiker to [prevent him from drinking the contaminated water and send him to camp elsewhere] [see if he usedthe contaminated water and if so, bring him out to seek medical attention]. However, if they send the search andrescue team after the hiker, they will be unavailable for other emergencies which may arise. Consider theseriousness of other problems you know about or can imagine that the search and rescue team may be called onto handle. Relative to these other emergencies, what priority would you place on sending the search and rescueteam after the hiker?

Figure 1. Logical relationships between the four possible causes. Arrows represent implication.



5% 50% 90% 100%


the within-subject comparisons of the direct comparison condition. All subjects receivedall versions of each scenario, with scenarios designed to test the same cause separated byother questions.

Results. Despite the heterogeneity of the three sample populations, there were no signifi-cant differences in their responses, and no order effects, so the data were aggregated. Theresults for both the rating and the direct comparison conditions are summarized in Table5. This table lists the number of respondents rating or choosing each scenario as moreimportant. For each of the four possible causes, the first column of data shows the number(fraction in parentheses) of subjects whose responses are consistent with the explanationbeing tested. For example, if the vividness of the victim is a cause of the identifiablevictim effect, subjects should say it is more important to take action when the victim isvividly described. The first column of data shows the number of subjects who ranked thescenarios in an order consistent with this explanation. The second column of data showsthe number answering in the direction inconsistent with the explanation, and the thirdcolumn shows the number who rated the two programs as equally important, or who chose“equally important” in the direct comparison version. The final column indicates whetherthere is a significant difference between the number of subjects responding in the pre-

Table 5. Number of subjects rating each scenario as more important: Study 1.

Vivid Anonymous

Equally

important Significance

VividnessRatings 8 (.06) 13 (.15) 64 (.79) NSDirect Comparison 4 (.10) 1 (.03) 35 (.88) NS

Certain UncertainEqually


Certainty and uncertaintyRatings 26 (.30) 6 (.07) 54 (.63) ,0.005Direct Comparison 11 (.29) 8 (.21) 19 (.50) NS

Smallerreference

group

Largerreference

groupEqually


Proportion of the reference groupRatings 44 (.51) 13 (.15) 29 (.34) ,0.001Direct Comparison 21 (.51) 3 (.07) 17 (.41) ,0.001

Ex post Ex anteEqually


Ex post/ex anteRatings 12 (.14) 24 (.28) 50 (.58) (1)Direct Comparison 7 (.17) 9 (.22) 25 (.61) NS

Numbers in parentheses show the fraction of respondents answering in each order.(1) p , 0.05, significant in opposite direction from prediction.



5% 50% 90% 100%


dicted direction and the number responding in the opposite direction. To determine sig-nificance, the number of subjects responding in the predicted direction was compared withthe number responding in the opposite direction using a sign test.

The modal response to most of the scenarios was to rate the two situations as equiva-lent. That is, for questions testing each possible explanation, subjects felt it was equallyimportant to save lives in the two scenarios. However, the responses of those subjectsexpressing a preference are quite interesting.

Subjects did not rate the familiar victim as more important than the anonymous victim,either in rating or direct comparison, contrary to the result that might be expected ifvividness is a primary determinant of the effect. There is no significant difference betweenthe responses in the rating and direct comparison conditions (x2(2) 5 4.49, p ' 0.11).

The rating task showed people to be significantly more concerned about certain thanabout uncertain deaths, but the direct comparison task did not. In this case, the differencebetween respondents in the direct comparison and the rating versions is marginally sig-nificant (x2(2) 5 5.47, p , 0.1).

Subjects are significantly more concerned with saving lives when they represent a largeportion of the reference group. In fact, for the proportion-of-the-reference-group ques-tions, the modal response was to rate saving 25 out of 25 casualties as more importantthan saving 25 out of 50,000. This was the case in both the rating and the direct com-parison versions, and further, there is no significant difference in the responses from thetwo conditions (x2(2) 5 1.8, p ' 0.4).

Finally, subjects in the rating condition felt it was more important to take action ex ante(preventative measures) than to take action ex post (remedial measures). This is theopposite of the result that is expected if the ex post/ex ante distinction is a cause of theidentifiable victim effect. However, the effect of this distinction was not significant in thedirect comparison condition. Note, however, that there is no significant difference betweenthe responses in the rating and direct comparison conditions (x2(2) 5 0.60, p ' 0.7).

Discussion. The results of this experiment provide support for two possible causes of theidentifiable victim effect: the certainty effect and the proportion-of-the-reference grouphypothesis. Furthermore, the differences between responses in the rating and direct com-parison conditions indicate that the effect of certainty appears unintended, whereas thepercentage reduction in risk is something that subjects consciously take into account.

We found no support for the hypothesis that the vividness with which a victim isdescribed increases subjects’ desire to save that victim. Apparently simply knowing that“a person” is definitely at risk and can be saved by taking action is enough to engender ourconcern.

Finally, the hypothesis that the distinction between ex post and ex ante action is criticalwas not supported by our results. To the degree that subjects care about this distinction,they appear to believe that preventative actions are more important than remedial actions.

The lack of response to the vividness manipulation is not easily explained, unless therewas something specific about the “vivid” victim that made her unsympathetic. However,in retrospect, there may be alternative explanations for some of our other results. Thequestions testing the certainty hypothesis are acknowledged to be highly unlikely, al-



5% 50% 90% 100%


though the two scenarios are similarly implausibile, so this alone would not be enough toproduce the effect found. However, a food distributor can be blamed for a contaminationproblem much more easily than for allergic reactions. The saliency of blame may havecontributed to the perceived importance of saving the more certain victims. The scenariostesting the proportion of the reference group hypothesis describe two programs, each ofwhich is estimated to save 25 lives. The program saving 25 out of 25 might naturally beseen as more effective than that saving 25 out of 50,000, or more believable, or morecost-effective. These reactions would result in subjects rating saving 25 out of 25 lives asmore important than saving 25 out of 50,000 lives. Finally, in the scenarios testing thedifference between ex post and ex ante action, the ex ante decision could be simply not toapply the pesticide to the field. The existence of an easy and costless solution to preventthe risk ex ante may account for the direction of the effect.

3.2. Study 2

Given the somewhat surprising results obtained from the first study—the failure to find avividness effect and the strong effect of reference group size—we designed a second studyto test further the same four possible causes of the identifiable victim effect. Again weused a variety of different scenarios, manipulating the vividness of the victim description,the degree of uncertainty about the risks, the proportion of the reference group at risk, andwhether the proposed action takes place before or after the risks are realized.

Method. Given the weak response to the various manipulations in Study 1, our first goalwas to validate the identifiable victim effect itself by comparing subjects’ reactions to anobvious example of an identifiable versus statistical fatality, where the identifiable casedemonstrates most or all of the four characteristics and the statistical case demonstratesnone. We developed two scenarios involving lead poisoning in children, presented in Table6. In both cases action could be taken that had a 0.1% chance of saving a life. In theidentifiable case the victim was a child who was hospitalized with acute lead poisoning,

Table 6. Scenarios testing the effect of identifiable and statistical victims.

Study 2

[Identifiable] Suppose that you are a hospital administrator running a large metropolitan hospital undertight budget constraints. A young child has been brought in with acute lead poisoning, and is unlikely to live.His physician has suggested trying new, untested treatment that might help. However, the treatment is experi-mental, very expensive, and is estimated to have only a 0.1% chance of saving the child’s life. Assume that youdecide not to approve the treatment, and that the child dies. How personally responsible would you feel?

[Statistical] Suppose that you are a hospital administrator running a large metropolitan hospital under tightbudget constraints. A local community group has requested that the hospital provide free lead level screeningtests to all children in the community. Comprehensive lead screening of all children would be very expensive,the experience with lead levels in the community suggest there is only a 0.1% chance any child will be exposedto fatal levels of lead under current circumstances. Assume that you decide not to institute the lead-levelscreening program, and later in the year a child dies from acute lead poisoning. How personally responsiblewould you feel?



5% 50% 90% 100%


and certain to die unless a new, experimental treatment is used (and likely to die even withthe treatment). So the victim is described (vivid), comprises 100% of the reference group,is certain die without treatment, and has already experienced the risk. In the statisticalcase the proposed action is preventative (community-wide lead screening tests), and thevictim is an anonymous child, comprising a small proportion of the reference group (allchildren in the community), who is not certain to be exposed or die from the risk, and whohas not yet experienced the risk. In both scenarios subjects were asked to assume that theydid not take action (did not approve the treatment or did not fund a testing program) anda child died. They were asked to rate how responsible they would feel, on a one to sevenscale. Since the victim in the identifiable scenario possesses all of the characteristics wehypothsize may cause the identifiable victim effect, we predicted that subjects will feelmore responsible for that death than for the death of one anonymous child in the com-munity.

To test the vividness hypothesis more thoroughly we developed a set of scenarios whichinclude a variety of specific victims, and different levels of description. In this scenario avictim is trapped in a collapsed structure after an earthquake and rescue efforts areunderway. Subjects read a realistic news story describing the situation, where the victimis presented either (1) generically, as “a man” or “a woman,” (2) with a descriptionincluding the sex, age, occupation, marital status, and quotes about the victim fromfriends or relatives, or (3) with the same description and a picture. After reading about thevictim, subjects were asked how strongly they would support (on a one to seven scale) avery expensive action (closing a nearby freeway) to reduce risks marginally to the trappedvictim. Two generic and eight specific victims were created, half male and half female.The ages and occupations of the victims were randomly selected using the base rates ofthe ages and occupations of U.S. citizens. Using a variety of victims should reduce thepossibility that subjects are simply responding to a very sympathetic (or unsympathetic)victim. The second panel of Table 1 presents the scenario and an example of one of thevictim descriptions. If the vividness of the description matters, subjects should be morewilling to close the freeway for the victims who are described, or described and pictured,than for generic victims. Subjects were also asked to indicate how much they cared aboutwhat happened to the trapped victim.

Uncertainty about expected fatalities can come from two sources: uncertainty aboutwho will be a victim, and uncertainty about how likely any given victim is to die. In Study1, we attempted to manipulate only the second type of uncertainty. In Study 2, wecombined these two types of uncertainty and presented a range of possible fatalities. Theuncertainty scenarios involved young children suffocating on thin-layer plastic wrap andbags. The scenarios are shown in Table 2. Subjects were told that the average number ofdeaths from suffocation for children under the age of 4 is 100 per year, with a range of 92to 110 (low variance case) or 34 to 168 (high variance case). They were asked to indicatehow strongly they would support legislation requiring the phasing out of thin-layer plasticin packaging and drycleaning. While in neither of these scenarios are fatalities certain, ifpeople are more concerned about “certain” fatalities, they should be more concernedabout the fatalities in the low variance case, which are “less uncertain,” than in the highvariance case. To test whether the possibility of zero fatalities carries any special signifi-



5% 50% 90% 100%


cance, we ran a second uncertainty study where the expected number of deaths was 20,and the uncertainty ranged from 15 to 25, or from 0 to 40.

To test the effect of the proportion of the reference group that can be saved, we againused a scenario involving traffic fatalities. These scenarios are presented in Table 3. Toreduce any ambiguity about the locations of the fatalities, and the type and efficacy of therisk-reducing actions, we described a specific type and location of accident (pedestrianfatalities at a single intersection in downtown Pittsburgh), as well as the proposed actionsto eliminate those risks (installing auto barriers and a pedestrian overpass). Only the sizeof the reference group was varied (2 of 4, 112, or 1700), and subjects were asked toindicate what priority they would assign to the proposed project on a one–to–ten scale.

Finally, the ex post/ex ante distinction was tested with a more believable scenario thanin the earlier study (see Table 4). A camper in the remote mountains either has beenexposed to dangerously contaminated water and is in need of immediate help (ex post), oris about to be exposed and in need of warning (ex ante)3. Subjects were asked to rate theimportance of rescuing this victim.

Subjects. The questionnaire was administered to 121 adult visitors to the PittsburghInternational Airport. All subjects answered only one question from each set of scenarios:that is, one question designed to test each of the hypothesized causes. Questions relatedto each hypothesis were randomized over 24 different versions of the survey. The secondquestion involving uncertainty was answered by 100 adult visitors to the airport.

Results. To reduce variance caused by inter-subject heterogeneity in average concern forvictims and in use of the scales, we normalized each subject’s responses by subtractingfrom each concern rating the weighted average of all that subject’s concern ratings. Table7 shows the mean normalized rating for each question. The last column indicates whetherthe difference in mean ratings reaches statistical significance, by either a t-test or an F-test,as appropriate. As shown in the table, most of the questions did not yield significantlydifferent mean ratings.

Subjects feel significantly more responsible for the specific, identifiable victim in themedical (lead-level) questions who dies because s/he doesn’t get treatment than for theanonymous victim who dies because the county-wide lead screening program is notapproved. This is simply the identifiable victim effect itself, demonstrated in an experi-mental setting.

We again observed an unexpected effect of vividness, even though we stengthened thevividness manipulation and took pains to ensure that victim described was statisticallyrepresentative. Although in Study 1 subjects were more concerned about the anonymousvictim, in this study subjects stated that they cared the most when a verbal description ofthe victim was provided and slightly less when the victim was described simply as “aman” or “a woman,” although the difference did not approach statistical significance.However, surprisingly, subjects cared least when the a picture of the victim was includedalong with the description. No consistent effect was observed of vividness on willingnessto support the risk-reduction project.



5% 50% 90% 100%


In a slight departure from our earlier results, Study 2 failed to find significant differ-ences in the importance ratings for the questions testing the effect of different levels ofuncertainty.

Consistent with the results of Study 1, however, subjects place significantly higherpriority on a project that is estimated to save two lives if those lives represent a highproportion of the reference group (2 out of 4), than on that identical project if those sametwo lives represent only a small proportion of the reference group (2 out of 1700).Although this effect is significant only at the .06 level (two-tailed test), the ANOVA isconservative because it does not take account of the ordering of means, which is as-predicted.

Finally, we did not find a significant effect for the ex post/ex ante distinction.

Discussion. The results of this experiment provide additional support for one possiblecause of the identifiable victim effect: the proportion of the reference group at risk appearsto be an important factor affecting subjects’ support for risk-reducing actions.

As in Study 1, these results do not support the hypotheses that more vivid descriptionsincrease subjects’ concern for, or desire to save, that victim, that people care more aboutmore certain than less certain fatalities, or that ex post victims are more important than ex

Table 7. Mean rating for each scenario in Study 2.

n Mean rating Significance

Identifiable/StatisticalIdentifiable 54 21.61 p , .005Statistical 60 22.40 [t(112)]

Vividness–support projectno description 38 1.12 NSdescription 41 .61 [F(2,111)]description with picture 35 1.03

Vividness–care about victimno description 38 .44 NSdescription 41 .64 [F(2,111)]description with picture 35 .17

Certainty and uncertaintylow variance (92–110) 58 .70 NShigh variance (34–168) 55 1.12 [t(112)]low variance (15–25) 49 5.90 NShigh variance (0–40) 51 5.23 [t(98)]

Proportion of the reference grouplarge (2 of 4) 39 .66 p , .06medium (2 of 112) 35 .19 [F(2,111)]small (2 of 1700) 40 2.25

Ex ante/ex postex ante 59 .74 NSex post 55 1.09 [t(112)]



5% 50% 90% 100%


ante victims. The uncertainty finding is a slight departure from the results of Study 1,where certain risks were judged more important than uncertain risks, although only in therating condition.

4. General discussion

Although there have been numerous references to the identifiable victim effect, this studyis, to the best of our knowledge, the first to examine the effect systematically. Despite thesuperficial simplicity of the distinction between identifiable and statistical lives, we notedthat there are actually several differences between them that could account for theirdifferential treatment.

One major surprise to emerge from the studies is that vividness does not appear to havean effect on subjects’ willingness to support risk–reducing actions. When we have spokento friends and colleagues about this project, many propose vividness as the explanation forthe identifiable victim effect. We should note, however, that our research is not the first toobtain weak vividness effects (Taylor and Thompson, 1982).

Based on our research, of course, we cannot conclude that all vividly described victimswill be seen as no more important than anonymous victims in terms of decision making.Certainly it would be possible to create scenarios with a particularly compelling or sym-pathetic victim, or a vivid scenario, in which subjects would express a preference forsaving the “familiar” victim. Our experiment attempted to see if more detailed informa-tion about the victim would, by itself, cause the identifiable victim effect. The questionsof what information about a victim increases our sympathy, and of what, if any, informa-tion will make us care more about one identified victim than about another identifiedvictim are interesting issues discussed in the literature on helping behavior (see, e.g.,Piliavin, Rodin and Piliavin, 1969), but not addressed by our study.

When victims are identified it is clear exactly how many people will die, but whenvictims are statistical it is always possible that more or fewer will die. In our first study,subjects felt avoiding certain fatalities was more important than avoiding uncertain fa-talities when they were not able to compare scenarios directly. However, subjects judgedcertain and probabilistic deaths as equally important when they compared the two situa-tions directly. Thus, the judgment that it is more important to address certain risks than toaddress probabilistic risks appears to be an unconscious one. Our second study found nosignificant difference between the reaction to high or low variance distributions when theexpected number of deaths is held constant.

Our findings suggest that the major cause of the identifiable victim effect is the relativesize of the reference group compared to the number of people at risk. Identified victimsconstitute their own reference group, 100% of whom will die if steps are not taken to savethem. Further, the response to the relative size of the reference group is consistent evenunder conditions where subjects explicitly contrasted scenarios which clearly differed onlyin the fraction of those at risk who can be saved. Thus consideration of the proportion ofthose at risk who can be saved appears to be a factor subjects would endorse for makingdecisions about risk-reducing activities.



5% 50% 90% 100%


The ex post/ex ante results are somewhat less surprising on reflection. It appears thatonce we know an individual is definitely at risk, there is no difference in the importanceof taking action after the risk is realized or before the risk has occurred: in fact, there isa slight preference for preventative action. Conventional wisdom suggests this result, aswe often hear “an ounce of prevention is worth a pound of cure.” Perhaps the ex post/exante distinction is simply a short-hand way for referring to the multitude of emotional andethical issues that come into play once a victim has been identified; isolated from thoseissues, it does not appear to produce the identifiable victim effect.

In combination, these results point to the somewhat surprising conclusion that theidentifiable victim effect, per se, may be wholly attributable to the effect of the relativesize of the reference group. We wonder whether the identifiable victim effect could moreaccurately (but less elegantly) be labeled the “percentage of reference group saved effect.”

If the identifiable victim effect is, in fact, largely due to the relationship betweenidentifiability and the size of the reference group, this raises significant questions aboutthe normative status of the effect and the role it should play in policy decisions, becausethe normative arguments for a reference group effect are tenuous. The reference group isoften largely a matter of framing, and it is difficult to defend a distinction between asituation where there is a group of 10 randomly distributed “vaccine sensitive” peoplewho are at risk of death from a flu vaccine, but who cannot be identified beforehand, anda situation in which 10 random people will be killed by the same vaccine.

Most policy decisions about risk involve statistical fatalities, while most private deci-sions involve identifiable fatalities. The normative status of the effect is not necessarilyrelevant to private decisions—no one would declare it irrational for parents to go to allextremes to save the life of their child. However, it is relevant to public policy deci-sions—we can legitimately ask whether it makes sense for society to go to extremes tosave one identified life when those resources could be spent more productively to save alarger number of statistical lives. As Keeney (1995) notes, there is no right or wronganswer. He suggests that we may want to assign different economic values to identifiedand statistical lives. But which of these values should form the basis for policy decisions?Allowing risk policies to vary depending on whether the victims are identified or statis-tical may create incentives for advocates of one policy to play up the identifiable victimsthat could be saved under that policy, while pointing out that we don’t know who will besaved under another. However, Viscusi (1992) points out that if we assign a higher valueto saving any identified victim than to saving a statistical victim, then perhaps we need torethink how we value statistical victims, since at some point all victims are identified.

Questions about whether and how identifiable and statistical victims should be consid-ered differently in policy decisions are not easily answered. However, given the arbitrari-ness of the reference group that applies to a specific risk, it seems inadvisable to recom-mend reference group size as an input into public policy except, perhaps, when the groupis defined geographically or by a sensitive demographic characteristic.

Why did the plight of young Jessica McClure engender such sympathies and such astrong response? Certainly she represented a very sympathetic victim, and the mediacoverage of the event ensured that we knew a great deal about her. It would seem heartlessto suggest that she not be saved, or that a cost-benefit analysis be conducted before rescue



5% 50% 90% 100%


efforts could commence. However, our study points to two other factors that may havebeen the most important in producing the powerful response: she was certain to die if notremoved from the well, and she comprised 100% of the risk group.

Acknowledgements

We thank Jonathan Baron, Graham Loomes, Keith Murnighan, Daniel Read, and PeterUbel for helpful comments. This work was supported under a National Science Founda-tion Graduate Research Fellowship, and by the Center for Integrated Study of the HumanDimensions of Global Change (NSF Grant #SBR 95-21914). Any opinions, findings,conclusions, or recommendations expressed in this paper are those of the authors and donot necessarily reflect the views of the National Science Foundation.

Notes

1. Although the identifiable victim effect may apply to many less severe, impacts, our focus in this paper is onfatal victims.

2. Scenario studies of this type have several limitations. First, the manipulated factors will almost inevitablyinteract with the specific content of the scenarios, raising questions about external validity (see Shotland,1983). Second, subjects rate their own level of concern, raising the issue of self-presentation and theaccuracy of introspection.

3. We thank an anonymous reviewer for suggesting this scenario.

References

Batson, C. Daniel, et al. (1991). “Empathic joy and the empathy-altruism hypothesis.” Journal of Personality andSocial Psychology 62, 413–26.

Bullard, Robert D. (1993). Confronting Environmental Racism: Voices from the Grassroots. Boston: South EndPress.

Calabresi, Guido, and Philip Bobbitt (1978). Tragic Choices. New York: W.W. Norton and Co.Cialdini, Robert B., et al. (1987). “Empathy-based helping: is it selflessly or selfishly motivated?” Journal of

Personality and Social Psychology 52, 749–58.Cohen, M, J. Y. Jaffray, and T. Said (1987). “Experimental comparison of individual behavior under risk and

under uncertainty for gains and for losses,” Organizational Behavior and Human Decision Processes 39,1–22.

Cushman, John H. Jr. (1994). “Clinton to Order Effort to Make Pollution Fairer,” The New York Times, February10, page A1.

Douglas, Mary (1992). Risk and Blame: Essays in Cultural Theory. New York: Routledge.Egan, Timothy (1994). “A 3-Strike Law Shows It’s Not as Simple as it Seems,” The New York Times, February

15, p. A1.Gibbard, Allan (1986). “Risk and Value,” in Douglas MacLean (Ed.), Values at Risk. New Jersey: Rowan and

Allanheld.Gillette, Clayton P., and Thomas D. Hopkins (1988). Federal Agency Valuations of Human Life. Administrative

Conference of the United States, Report for Recommendation 88–7.Glover, Jonathan (1977). Causing Death and Saving Lives. New York: Penguin Books.



5% 50% 90% 100%


Goodman, Walter (1993). “TV, by its very nature, can stack the deck,” The New York Times, September 13, p.C20.

Gore, Al (1992). Earth in the Balance: Ecology and the Human Spirit. New York: Plume.Horowitz, John K. and Richard T. Carson (1993). “Baseline risk and Preference for Reductions in Risk to Life.”

Risk Analysis 13 (2), 457–462.Keeney, Ralph L. (1995). “Understanding Life-Threatening Risks,” Risk Analysis 15 (6).Kahneman, Daniel and Amos Tversky (1979). “Prospect Theory: An Analysis of Decision Under Risk.” Econo-

metrica 47(2).Latané, Bibb, and John M. Darley (1970). The Unresponsive Bystander: Why doesn’t he help?. New York:

Appleton-Century-Crofts.Linden, Eugene (1988). “Helping Out Putu, Siku and Kanik,” Time Magazine, pp. 76–77. October 31.MacLean, Douglas (1986). “Social Values and the Distribution of Risk,” in Douglas MacLean (Ed.), Values at

Risk. New Jersey: Rowan and Allanheld.National Research Council (1989). Improving Risk Communication. Washington DC: National Academy Press.Nisbett, Richard and Lee Ross (1980). Human Inference: Strategies and Shortcomings of Social Judgment. New

Jersey: Prentice-Hall, Inc.People Weekly (1987). “America’s heart goes out to Baby Jessica.” Volume 28:18. November 2.People Weekly (1990). Volume 33:15. April 16.Piliavin, Jane A., et al. (1981). Emergency Intervention. New York: Academic Press.Piliavin, Irving M., Judith Rodin, and Jane A. Piliavin (1969). “Good samaritanism: an underground phenom-

enon?” Journal of Personality and Social Psychology 13, 289–299.Rawls, John (1971). A Theory of Justice. Cambridge, Massachussetts: Harvard University Press.Rawls, John (1993). Political Liberalism. New York: Columbia University Press.Raynor, Steve (1992). “Cultural Theory and Risk Analysis,” In Seldon Krimsky and Dominic Golding (Eds.),

Social Theories of Risk. Connecticut: Praeger Press.Redelmeier, Donald A. and Amos Tversky (1990). “Occasional Note: Discrepancy Between Medical Decisions

for Individual Patients and for Groups,” The New England Journal of Medicine, April 19.Ritov, Ilana, and Jonathan Baron (1990). “Reluctance to Vaccinate: Omission Bias and Ambiguity,” Journal of

Behavioral Decision Making 3, 263–277.Schelling, T. C. (1968). “The Life You Save May Be Your Own,” in Samuel Chase (Ed.), Problems in Public

Expenditure Analysis. Washington DC: The Brookings Institute.Shotland, R. Lance (1983). “What’s wrong with helping behavior research? Only the independent and dependent

variables.” Academic Psychology Bulletin 5, 339–350.Shrader-Frechette, K. S. (1991). Risk and Rationality: Philosophical Foundations for Populist Reforms. Berke-

ley: University of California Press.Skelton, George (1993). “A Father’s Crusade Born From Pain,” The Los Angeles Times, December 9, p. A3.Slovic, P., B. Fischhoff, and S. Lichtenstein. (1980). “Facts and Fears: Understanding Perceived Risk,” In Richard

C. Schwing and Walther A. Albers, Jr. (Eds.), Societal Risk Assessment: How Safe is Safe Enough? New York:Plenum Press.

Taylor, S. E. and S. C. Thompson (1982). “Stalking the elusive ‘vividness’ effect.” Psychological Review 89,155–181.

Toufexis, Anastasia (1993). “The Ultimate Choice.” Time Magazine, pp. 43–44. August 31.Tversky, Amos, and Daniel Kahneman (1981). “The Framing of Decisions and the Psychology of Choice,”

Science 211, 243–258.Tversky, Amos, and Daniel Kahneman (1986). “Rational Choice and the Framing of Decisions,” Journal of

Business 59 (4), part 2.Variety (1989). “TV Reviews—Network: Everybody’s Baby,” Volume 335:7, May 31.Viscusi, W. Kip and William N. Evans (1990). “Utility Functions That Depend on Health Status: Estimates and

Economic Implications,” The American Economic Review 80 (3), 353–374.Viscusi, W. Kip (1992). Fatal Tradeoffs: Public and Private Responsibilities for Risk. New York: Oxford Uni-

versity Press.



5% 50% 90% 100%


Weinstein, Milton C., Donald S. Shepard, and Joseph S. Pliskin (1980). “The Economic Value of ChangingMortality Probabilities: A Decision-Theoretic Approach,” Quarterly Journal of Economics 94, 373–396.

Whipple, Chris (1992). “Inconsistent Values in Risk Management” in Seldon Krimsky and Dominic Golding(Eds.), Social Theories of Risk. Connecticut: Praeger Press.



5% 50% 90% 100%

Explaining the “Identiﬁable Victim Effect” · explaining the “identifiable victim effect”

Documents

Transcript of Explaining the “Identiﬁable Victim Effect” · explaining the “identifiable victim effect”